Handwritten Text Extraction from Bank Cheque Images by a Multivariate Classification Process

摘要

In this paper we present an identification and extraction process of textual elements based on some local features. Using local features produces a document independent process, which is therefore more generic and free of heuristic based decisions, as in the case of contextual information that is also discarded in this work. The observed features supply a classifier which distinguishes between handwritten and machine printed elements. We detail some shape and content based features and present a set of them elected to perform the textual elements classification. Bank cheque images from several different institutions and writers were used to evaluate the performance of the process. Preliminary results demonstrate the efficiency of this approach and its potential. A brief comparison with previous works is also presented in order to highlight some benefits of the present methodology.